Professional Documents
Culture Documents
INTRODUCTION
Pint erweaving (t )
PAPER ORGANIZATION
Rest of the paper is organized as follows: In section II we
provide the design of our MAC protocol. We discuss about
the pros and cons of interweaving and underlay approaches.
Further we give a detailed comparison between full and half
duplex modes in this section. The details of our proposed
MAC protocol and mathematical analysis are elaborately
discussed in section III. Performance evaluation of the
proposed design is shown in section IV. Section V discusses
the future research opportunities and thereafter concludes the
paper.
II.
I allowable
A. Network Model
We consider a single channel where the duration of OFF
period of PU follows a probability distribution with
probability density function (pdf), A 0 (t ) and cumulative
distribution function (cdf), C 0 (t ) . It is already shown in
empirical spectrum measurement data [9], [10] that heavytailed or exponentially tailed distributions are the most
appropriate to approximate PUs ON/OFF duration. Thus we
assume that the ON and OFF period of PU, TON and TOFF are
approximately known to SUs (Refer to Appendix A).
B. Hybrid Full-/Half-Duplex MAC Protocol
Table 1: Notation
Symbol
Pd and Pf
Meaning
Probability of detection and probability of false
alarm respectively
Pd
fs
Sampling frequency
Mitigation coefficient
u and i
Sensing time
Td
Data slot
01 and10
B
g ( s 2 p ) (t )
Punderlay (t )
(1)
H1: y ( n) = k s ( n) + u ( n)
(2)
Where u(n) represents the zero-mean additive white
Gaussian noise (AWGN) with variance u2 and k is the
residual self-interference mitigation coefficient; 0 < k 1 ; k
describes the effect of self interference on full duplex
communication [11]. When k 0 , the effect of interference is
large on signal y(n) received at SU. And when k 1 , the effect
of interference is lowered. As in the case of half-duplex
communication, there is no scope for self-interference, we can
consider k=1 for half-duplex mode; whereas, k<1 for fullduplex mode.
The general expressions for Pd and Pf can be given as follows
[10]:
f s
Pd ( , k , ) = Q 2 k 1
k + 1
u
(3)
Pf ( , k , ) = Q 2 1 f s
(4)
Pf ( k , ) = Q ( k + 1)Q 1 ( Pd ) + k f s
(5)
Where, Td =
(6)
( )
B HD = Td 1 Q ( + 1) Q 1 Pd + T (1 Td ) f s
B FD = 1 Q ( k + 1) Q 1 ( Pd ) + k Tf s
))
(7)
(8)
1 Q ( k + 1)Q 1 ( Pd ) + k Tf s
1 Q ( + 1)Q 1 ( Pd ) + T (1 Td ) f s
)<T
(9)
(10)
Lemma 1: The inequality stated in eq. (10) holds for standard
values of parameters if k 2 > 1 Td .
Proof: Q1 ( x) < 0, if x > 1 2
> 0, if x < 1
( (
B0 = 1 Q ( k i + 1)Q 1 ( Pd ) + k i (TOFF + 01 ) f s
))
(12)
Again, for half-duplex underlay mode, from Fig. 1,
T = TON 01 + 10
(13)
And DT time = TON 01
(14)
Therefore, Td =
TON 01
TON 01 + 10
(15)
( (
))
TON 01
1 Q ( u + 1)Q 1 ( Pd ) + u 10 f s
(16)
B1 =
TON 01 + 10
Now, let B( 01,10 ) be the average normalized throughput as a
function of 01 and 10 .
Then, from eq. (12) and eq. (16), the optimization problem
can be formulated as:
Max : B( 01,10 ) = 1 [B0 ( 01,10 ) + B1 ( 01,10 )]
2
01,10
1 < Td
s.t .
(17)
( )
4. if L j < min {L ( )} do
( )
(i) min {L ( )} L j
(ii) * j
(iii) i 1
5. else compute Bcurrent
j *
*
6. if Bcurrent > Bbest do Bbest Bcurrent ; 01
01
;10 10j
(17.2)
7. endif
8. endif
9. if i> I max (upper limit)
(17.3)
(i)
01
Pint erweaving (t ) g
(s2 p)
(t ) dt I allowable
(17.4)
0
TOFF + TON + 10
underlay
Where,
( )
j
3. Solve L j using j as set of multipliers and obtain 01
,10j
(17.1)
10
TOFF +
TON 01 + 10
01
1 =
( t ) g ( s 2 p ) ( t ) dt u2
(17.5)
1 Q ( k i + 1)Q 1 ( Pd ) + k i (TON 01 + 10 ) f s
1 Q ( i + 1)Q ( Pd ) + i 10 f s
10
k
(TOFF + 01) f s
+
ON
01
10
2 = Q
1 k
(18)
) ( )( )
L * Bcurrent
|| ( min ) || t j t min
||
then STOP.
j
*
( = error tolerance)
12. endif
13. New subgradient:
(i) 1j = Td 1j Pd 2j
(ii) 2j = k 2
(ii) j
(iii) i0
(iv) goto step 2.
10. endif
11. if (j> J max (upper limit))
10
+ 1 (Td 1 ) Pd 2 + 2 k 2
ON
01
10
)(
10j
j
TON 01
+ 10j
( )
L j Bcurrent
|| j ||2
(iv) 1j +1 = 1j + t j 1j
(v) 2j +1 = 2j + t j 2j
14. goto step 2 while i I max and j J max .
15. end do-while
IV.
SIMULATION RESULTS
Parameters
Values
fs
Pd
0.9
6 MHz
Td
0.8
1 ms
Noise level
-95.2 dBm
0.2
0.8
Fig. 5 shows the comparison of our proposed hybrid half-/fullduplex MAC protocol with pure full-duplex and pure halfduplex MAC in terms of throughput. We observe that in the
low SNR regime, half-duplex MAC results into better
throughput than full-duplex MAC. But as SNR increases,
normalized throughput of full-duplex MAC rapidly increases
and performs better than half-duplex MAC. On the other
hand, our proposed hybrid MAC always provides higher
throughput than other two MAC protocols throughout the
whole SNR range [-25dB, -5dB]. This clearly shows why it is
more beneficial to use hybrid MAC instead of only full or only
half-duplex mode of communication and therefore, aptly
justifies the efficiency of our proposed MAC protocol in this
paper.
V.
CONCLUSION
= 0
L( )
=
0
10j
L( j )
j
01
j
Solving above system of differential equations by RungeKutta method, we get, 01j and 10j as functions of 1j , 2j .
j
01
= f (1j , 2j )
j
10 = g (1j , 2j )
POFF (t ) =
[6]
[7]
0 (t )dt
0 (t )dt
A (t)dt
C 0 (t )
C 0 (t ) + C 1(t )
(19)
[8]
C 1 (t )
C 1 (t ) + C 0 (t )
(20)
Again, when for t = TON , PON (TON ) is greater than a userdefined minimum value, we consider TON to be PUs ON
duration. In this paper, we have mainly considered three
distributions: (i) Gamma distribution, (ii) Lognormal
distribution and (iii) Pareto distribution. Detail procedure of
estimating PUs ON/OFF duration can be found in [6].
APPENDIX B: HOW TO SOLVE L( j ) AT j th ITERATION
IN STEP 3 OF ALGORITHM 1
By taking partial derivatives of L( j ) w.r.t 01j and 10j
and setting them equal to zero, we get,
(22)
REFERENCES
(21)
[9]
[10]
[11]
[12]
[13]